Skip to content

gpud-v0.8.0

Choose a tag to compare

@github-actions github-actions released this 07 Nov 01:14
· 242 commits to main since this release
2cf20de

GPUd release notes (2025-11-07T01:16:30Z)

Welcome to this new release!

What's Changed

  • fix(install.sh): add /etc/lsb-release fallback by @gyuho in #1064
  • feat: refactor dockerfile and expose additional helm chart variables by @MaxFedotov in #1068
  • feat(components/library): match more nvidia lib paths by @gyuho in #1069
  • [LEP-1758] feat(nvidia/xid): remove unused fields, new Xids for GB200+, rename packages, remove "ibstat" utils by @gyuho in #1061
  • [LEP-1877] feat(nvidia/bad-envs): remove component by @gyuho in #1053
  • [LEP-1740] feat(join): remove deprecated "join" command (use "gpud up" instead) by @gyuho in #1046
  • [LEP-1875] feat(nvidia/gsp-firmware-mode): degrade if supported but not disabled by @gyuho in #1054
  • feat(components): remove SetHealthy event writes by @gyuho in #1070
  • feat(go.mod): bump up nvidia nvlib, nvml deps by @gyuho in #1062
  • fix(os): do not create pstore store if directory does not exist by @gyuho in #1066
  • [LEP-1874] [LEP-2000] feat(components, disk): retain/lookback events 2 weeks by @gyuho in #1058
  • [LEP-2004] fix(disk): "findmnt" fallback for missing fstype, increase test coverage by @gyuho in #1059
  • [LEP-2083] fix(session): prevent deadlock waits on broken connection by @gyuho in #1065
  • [LEP-1325] [LEP-1736] feat(disk): remove "no space left" event, add health state checks based on free space left by @gyuho in #1045
  • [LEP-2000] feat(xid,sxid,gpu-counts): enable event store purge after 14 days by @gyuho in #1071
  • feat(nvidia/xid): allow to set reboot threshold (default remains 2) by @MaxFedotov in #1076
  • feat(server,cmd): add set healthy states local API endpoint, add "gpud set-healthy" command by @gyuho in #1073
  • chore: compatible with DGX Spark by @eahydra in #1081
  • [LEP-2343] fix(nfs): prioritize non-autofs mount over autofs mount by @gyuho in #1080
  • [LEP-2387] feat("accelerator-nvidia-gsp-firmware"): remove the component by @gyuho in #1083
  • [LEP-2083] fix(session): async triggerComponent response writes to avoid blocking the main serve loop by @gyuho in #1082
  • fix(nvidia/memory): handle getMemory "Not Supported" error gracefully by @gyuho in #1084
  • fix(machine-info): handle ERROR_NOT_SUPPORTED for GPU functions on CPU-only machines by @gyuho in #1090
  • [LEP-2600] fix(peermem,disk,*): run bash inline (without writing to disk) by @gyuho in #1089
  • chore: add Origin header by @eahydra in #1091
  • fix(set-healthy): allow to provide empty components list, fix incorrect response struct in the client by @MaxFedotov in #1093
  • fix(infiniband): remove unused int conversion helper by @gyuho in #1097
  • fix(nvidia/xid,sxid): fix panic in gpud scan by @gyuho in #1096
  • feat(xid,sxid): remove unused "CriticalErrorMarkedByGPUd" field by @gyuho in #1094
  • chore: support gpud-gateway by @eahydra in #1098
  • [LEP-2644] fix incorrect parsing of containerd version string by @jeking3 in #1099
  • chore: deprecated status and error fields from login response by @eahydra in #1102
  • nits(gpud/status): add file name to logging by @gyuho in #1100
  • doc(README): clarify resource requirements for Lepton Platform by @gyuho in #1101
  • [LEP-2600] feat(containerd): do not mark unhealthy when container service has just started (threshold 5-minute) by @gyuho in #1103
  • [LEP-2823] fix(nvidia/fabric-manager): auto-detect GB200 product, fix pci bridge detection for GB200 by @gyuho in #1105
  • [LEP-2823] fix(nvidia/fabric-manager): GB200 to skip "fabric-manager" activeness in favor of fabric state (optional) by @gyuho in #1106
  • feat(gpud/cmd): remove logout command by @gyuho in #1109
  • [LEP-2899] feat(nvidia/fabric-manager): use NVML APIs to monitor fabric state by @gyuho in #1108
  • [LEP-2957] fix(pkg/session): make process gossip async by @gyuho in #1113
  • [LEP-2957] feat(disk): set timeouts for GetPartitions operations by @gyuho in #1115
  • feat(nvidia/xid): add metric "accelerator_nvidia_xid_errors_total" by GPU UUID + XID by @MaxFedotov in #1116
  • nits(components/nfs): log health state transitions for debugging by @gyuho in #1117
  • feat(components/disk): remove unnecessary context timeout setups, improve logging around timeouts by @gyuho in #1119
  • feat(components/nvidia/xid): clarify event type by @gyuho in #1118
  • Removing Dead Links and Fixing Typo by @tucker-wilde in #1137
  • fix(nvidia/sxid,xid): fix nil pointer error in scan summary print by @gyuho in #1134
  • docs(versions): bump up to latest v0.8.0 by @gyuho in #1079

New Contributors

Full Changelog: v0.7.1...v0.8.0